Proceedings of MATSUS Spring 2024 Conference (MATSUS24)
DOI: https://doi.org/10.29363/nanoge.matsus.2024.162
Publication date: 18th December 2023
Recent advancements in Organic Solar Cells (OSCs) have led to peak efficiencies of up to 19%, primarily due to the development of novel organic molecules for active and transport layers. Despite this progress, OSC technology still encounters commercialization challenges. These are largely attributed to the extensive range of potential materials and the intricate interplay between their structures and properties, which complicates efficiency predictions. Addressing this, the materials science field is increasingly utilizing machine learning (ML) to decipher structure-property correlations.
Our research addresses this challenge by exploring the relationship between the electrical properties of OSCs, such as Power Conversion Efficiency (PCE) and Fill Factor (FF), and their structural features, including band center position and crystallinity. These are determined through the deconvolution of UV-Visible (UV-VIS) spectroscopy data. By employing Gaussian Process Regression, we have devised a method that predicts solar cell efficiency using only UV-VIS spectra. This approach potentially eliminates the need for evaporation and IV measurements, streamlining the process and reducing costs.
To ensure predictive accuracy, we compiled a dataset of 25 donor-acceptor combinations, yielding a total of 200 cells with intentional variations in film thickness and crystallinity. We ensured exceptional homogeneity in layer preparation using the Autonomous Materials and Device Application Platform (Amanda) 1. Analysis of this dataset enables us to foresee the performance of new material combinations. This significant breakthrough in OSC research could expedite the development of efficient and commercially viable solar cell materials.